log_parsers: keep cargo should_panic and doctest results

[k21993]

Summary

• Parse libtest result lines whose names contain a test mode or spaces (tests::x - should panic, src/lib.rs - add (line 30) - compile fail), in both log_parsers/cargo_parser.py and the standalone copy in _pr_runtime_verifier.py. On main, these lines don't match, so a fix whose new test is #[should_panic] has no FAIL_TO_PASS test and gets dropped.
• Key doctests by file and item without (line N), and let the worst status among an item's doctests win. The line number moves whenever a patch edits code above a doctest, so a verbatim key would score a correct patch that shifts lines as a regression.
• Match libtest's literal ... separator instead of \s+\.\.\.\s+, so names can contain spaces without backtracking on long lines.

The reproduction from #138 uses real cargo test output. The reference fix keeps the doctest at (line 3). The agent's patch is equally correct but adds a helper above div, which moves the doctest to (line 9).

Parser	F2P	P2P	Reference fix graded	Agent patch graded
main	[]	divides	task dropped	task dropped
spaces allowed, names verbatim	rejects_zero	divides, src/lib.rs - div (line 3)	1.0	0.5, doctest reported as a regression
this PR	rejects_zero	divides, src/lib.rs - div	1.0	1.0

Test plan

• uv run --all-extras pytest -q: 1912 passed, 6 skipped (live-network tests)
• uv run ruff check . and uv run ruff format --check .
• New tests/test_cargo_test_names.py runs every case against both parsers, the same way tests/test_pytest_node_ids.py does. Its fixtures are verbatim cargo test --no-fail-fast output from rustc 1.98.1: one crate covering every test mode, plus the pre-fix, reference-fix and agent runs above.
• 25 of the 40 new tests fail on main, and all 40 pass with this change.
• The standalone verifier, run with python -I, grades the shifted-doctest agent run at reward 1.0 with resolved=True.
• Four 200k-character pathological lines parse in a few milliseconds, with a subprocess timeout guard like the one in Fix pytest parsing of parametrized node IDs containing spaces #91.

Out of scope

• Tests with the same name in different crates or test binaries still collide, and the last write wins, as before.
• cargo nextest, --format terse or JSON output, and ANSI-colored logs.
• Tasks generated before this change keep the verifier baked into them.

Closes #138

[adithya-s-k]

Thanks, tested and merged. The should_panic and shifted-doctest cases now work in both parsers. Existing Cargo tasks will need their verifier and test IDs updated together.